Pierre-Antoine Laharotte
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Research Subjects
Pierre-Antoine Laharotte has a total of 27 co-authors. Here are the top co-authors by number of shared publications:
- Nour-Eddin El Faouzi
Univ. Gustave Eiffel, Univ. Lyon, ENTPE, LICIT-ECO7 UMR T9401
11 shared publications - Kinjal Bhattacharyya
Traffic Analysis and Logistics, Swedish National Road and Transport Research Institute (VTI), Linköping, Sweden
5 shared publications - Hugues Blache
Univ. Gustave Eiffel, Univ. Lyon, ENTPE, LICIT-ECO7, UMR T9401, Lyon, France
3 shared publications Nour-Eddin El Faouzi
3 shared publications- Philippe Bon
Université Gustave Eiffel
3 shared publications - Rim Brahim
Univ Gustave Eiffel, COSYS-ESTAS
3 shared publications
Latest Publications
- Open Access
- 12 November 2026
- Published by Tsinghua University Press in Journal of Intelligent and Connected Vehicles
Reinforcement Learning Agent Embedded in Roadside Units to Implement an Adaptive Variable Speed Limit System in a Cooperative Intelligent Transportation Systems Environment
Freeway shockwaves remain a major source of recurrent congestion, yet most reinforcement learning (RL) approaches to variable speed limit (VSL) control remain theoretical, relying on fixed detectors, predefined congestion zones and full observability. This paper introduces an RL agent embedded in a Roadside Unit (RSU) within a Cooperative Intelligent Transportation System. Operating solely through V2I communication, the agent collects speed and position data from Cooperative Awareness Messages, reconstructs traffic state as a space-time matrix built from sparse connected vehicle (CV) observations, and broadcasts speed limits, requiring no additional sensing infrastructure. Training proceeds in three stages: exhaustive dataset generation in a microscopic simulator, reward selection and hyperparameter calibration on a learned surrogate model, and few-shot refinement in the simulator. The transfer operates along two axes: (i) increasing environment fidelity, and (ii) the release of the simulation seed, so that a policy acquired deterministically is refined under stochastic conditions in 400 episodes, against the 4,000 required from a random policy. Across 30 seeds, the agent selects a consistent mitigation strategy from 30% CV penetration, where attenuation is partial; the physics-based benchmark performs better above 50%, while at full penetration the two are statistically indistinguishable, with lower inter-seed variance for the RL agent. Two coordinating RSUs with 2 km coverage manage a 10 km corridor.
- 12 November 2026
- Published by Tsinghua University Press in Journal of Intelligent and Connected Vehicles
Open AccessReinforcement Learning Agent Embedded in Roadside Units to Implement an Adaptive Variable Speed Limit System in a Cooperative Intelligent Transportation Systems Environment
Freeway shockwaves remain a major source of recurrent congestion, yet most reinforcement learning (RL) approaches to variable speed limit (VSL) control remain theoretical, relying on fixed detectors, predefined congestion zones and full observability. This paper introduces an RL agent embedded in a Roadside Unit (RSU) within a Cooperative Intelligent Transportation System. Operating solely through V2I communication, the agent collects speed and position data from Cooperative Awareness Messages, reconstructs traffic state as a space-time matrix built from sparse connected vehicle (CV) observations, and broadcasts speed limits, requiring no additional sensing infrastructure. Training proceeds in three stages: exhaustive dataset generation in a microscopic simulator, reward selection and hyperparameter calibration on a learned surrogate model, and few-shot refinement in the simulator. The transfer operates along two axes: (i) increasing environment fidelity, and (ii) the release of the simulation seed, so that a policy acquired deterministically is refined under stochastic conditions in 400 episodes, against the 4,000 required from a random policy. Across 30 seeds, the agent selects a consistent mitigation strategy from 30% CV penetration, where attenuation is partial; the physics-based benchmark performs better above 50%, while at full penetration the two are statistically indistinguishable, with lower inter-seed variance for the RL agent. Two coordinating RSUs with 2 km coverage manage a 10 km corridor.
- Open Access
- 17 October 2026
- Published by Tsinghua University Press
Reinforcement Learning Agent Embedded in Roadside Units to Implement an Adaptive Variable Speed Limit System in a Cooperative Intelligent Transportation Systems Environment
This data repository is related to the journal paper entitled "Reinforcement Learning Agent Embedded in Roadside Units to Implement an Adaptive Variable Speed Limit System in a Cooperative Intelligent Transportation Systems Environment", authored by Eléonore Fauchet (first author), Pierre-Antoine Laharotte, and Nour-Eddin El Faouzi. The study uses no field, proprietary or personal data. All data are synthetic and are generated by the microscopic traffic simulations described in the paper (SUMO). The package, therefore, contains the complete data, together with the scripts that regenerate them. The files provided are: – SUMO scenario files (XML); – Simulated datasets (CSV); – Trained models (Keras/HDF5, .h5); – Simulation outputs (CSV/XML). Because the trained weights and datasets are included, the evaluation results (Figs. 5–8) can be reproduced without re-running the training stages; the training stages can also be re-run from scratch with the scripts.
- 17 October 2026
- Published by Tsinghua University Press
Open AccessReinforcement Learning Agent Embedded in Roadside Units to Implement an Adaptive Variable Speed Limit System in a Cooperative Intelligent Transportation Systems Environment
This data repository is related to the journal paper entitled "Reinforcement Learning Agent Embedded in Roadside Units to Implement an Adaptive Variable Speed Limit System in a Cooperative Intelligent Transportation Systems Environment", authored by Eléonore Fauchet (first author), Pierre-Antoine Laharotte, and Nour-Eddin El Faouzi. The study uses no field, proprietary or personal data. All data are synthetic and are generated by the microscopic traffic simulations described in the paper (SUMO). The package, therefore, contains the complete data, together with the scripts that regenerate them. The files provided are: – SUMO scenario files (XML); – Simulated datasets (CSV); – Trained models (Keras/HDF5, .h5); – Simulation outputs (CSV/XML). Because the trained weights and datasets are included, the evaluation results (Figs. 5–8) can be reproduced without re-running the training stages; the training stages can also be re-run from scratch with the scripts.
- Open Access
- 13 October 2026
- Published by Tsinghua University Press in Journal of Intelligent and Connected Vehicles
Assessing Microscopic Traffic Simulation Confidence Level Through Expert-Weighted AHP and Belief Theory: An Application to CAV Testing
This paper addresses the challenge of assessing experts' confidence in simulation tools for reproducing safety-related indicators in traffic scenario-based testing. While simulation offers major advantages in terms of safety, cost, and repeatability, its outputs do not equally reflect reality across all traffic situations. The objective is therefore to identify which scenarios can be considered high confidence in simulation. The proposed approach relies on several Analytic Hierarchy Process (AHP) approaches combined with an expert survey on microscopic traffic simulation. Scenarios are represented through ontology-based attributes, and confidence weights are assigned through pairwise comparisons. Three fusion strategies are compared: classical AHP, weighted AHP based on expertise level, and AHP combined with belief theory. The results show that accounting for expertise improves the confidence scores and leads to more consistent rankings of scenarios. Dynamic traffic and environmental conditions are generally associated with lower confidence than static road-related attributes. Scenario score distributions and Gaussian mixture analysis further reveal distinct groups of low-, medium-, and high-confidence scenarios. These results provide a decision-support framework for distinguishing scenarios associated with higher expert-assigned confidence from those that may require real-world testing.
- 13 October 2026
- Published by Tsinghua University Press in Journal of Intelligent and Connected Vehicles
Open AccessAssessing Microscopic Traffic Simulation Confidence Level Through Expert-Weighted AHP and Belief Theory: An Application to CAV Testing
This paper addresses the challenge of assessing experts' confidence in simulation tools for reproducing safety-related indicators in traffic scenario-based testing. While simulation offers major advantages in terms of safety, cost, and repeatability, its outputs do not equally reflect reality across all traffic situations. The objective is therefore to identify which scenarios can be considered high confidence in simulation. The proposed approach relies on several Analytic Hierarchy Process (AHP) approaches combined with an expert survey on microscopic traffic simulation. Scenarios are represented through ontology-based attributes, and confidence weights are assigned through pairwise comparisons. Three fusion strategies are compared: classical AHP, weighted AHP based on expertise level, and AHP combined with belief theory. The results show that accounting for expertise improves the confidence scores and leads to more consistent rankings of scenarios. Dynamic traffic and environmental conditions are generally associated with lower confidence than static road-related attributes. Scenario score distributions and Gaussian mixture analysis further reveal distinct groups of low-, medium-, and high-confidence scenarios. These results provide a decision-support framework for distinguishing scenarios associated with higher expert-assigned confidence from those that may require real-world testing.
- Open Access
- 28 September 2026
- Published by Springer Nature in European Transport Research Review
- Vol. 18 (1) , 89
Level crossing traffic management for an innovative on-demand public transport rail system
Managing priority at intersections where conventional road vehicles interact with automated light rail systems poses new challenges, particularly with the emergence of Ferromobile. Ferromobile is an autonomous public transport service operating on existing railways, either through fixed schedules or demand-responsive transport (DRT). By reactivating disused rural lines in France, it aims to improve regional mobility. Unlike conventional trains, Ferromobiles are similar in size and weight to road shuttles, raising specific operational concerns at level crossings (LCs). The irregular arrivals inherent to on-demand services make it problematic to grant systematic priority to Ferromobiles, as this may lead to congestion and unsafe driver behavior. This calls for a redefinition of priority rules between rail and road traffic, contrary to conventional railway practice. We propose an optimization-based framework that uses a moving time window to dynamically allocate passage slots to Ferromobiles while respecting safety constraints and traffic throughput objectives. The objective function pursues two goals: (i) reducing rail delays and (ii) minimizing disruption to road traffic. In this article, two optimization models, CQCM and PQCM, were implemented on the Pyomo platform and compared both with each other and with two reference scenarios without optimization in order to evaluate their operational efficiency and impact on LC traffic management.
- 28 September 2026
- Published by Springer Nature in European Transport Research Review
- Vol. 18 (1) , 89
Open AccessLevel crossing traffic management for an innovative on-demand public transport rail system
Managing priority at intersections where conventional road vehicles interact with automated light rail systems poses new challenges, particularly with the emergence of Ferromobile. Ferromobile is an autonomous public transport service operating on existing railways, either through fixed schedules or demand-responsive transport (DRT). By reactivating disused rural lines in France, it aims to improve regional mobility. Unlike conventional trains, Ferromobiles are similar in size and weight to road shuttles, raising specific operational concerns at level crossings (LCs). The irregular arrivals inherent to on-demand services make it problematic to grant systematic priority to Ferromobiles, as this may lead to congestion and unsafe driver behavior. This calls for a redefinition of priority rules between rail and road traffic, contrary to conventional railway practice. We propose an optimization-based framework that uses a moving time window to dynamically allocate passage slots to Ferromobiles while respecting safety constraints and traffic throughput objectives. The objective function pursues two goals: (i) reducing rail delays and (ii) minimizing disruption to road traffic. In this article, two optimization models, CQCM and PQCM, were implemented on the Pyomo platform and compared both with each other and with two reference scenarios without optimization in order to evaluate their operational efficiency and impact on LC traffic management.
- 3 April 2026
- Published by Agence Bibliographique de l'Enseignement Supérieur
Contributions à la prévision court-terme, multi-échelle et multi-variée, par apprentissage statistique du trafic routier
La maturité de la télématique et des technologies de l’information et la communication (TIC), ainsi que l’avènement du big data dans le transport ont conduit à des développements foisonnants dans le domaine des systèmes de transports intelligents (ITS), aussi bien sur le plan des technologies de recueil que du traitement innovant de l’information. Il est désormais possible de connaître les conditions de circulation et les états de trafic sur la plupart des sections d’un réseau routier sans avoir recours à des infrastructures intrusives de collecte de données, de transmettre l’information résultante via des réseaux sans fil et de traiter rapidement toutes ces données multi-sources disponibles. La constitution de grandes bases de données a naturellement fait évoluer la pratique de gestion du trafic et plus particulièrement les méthodes de prévision. Ces méthodes ont connu un renouveau en s’inspirant des travaux produits en apprentissage statistique. Néanmoins, la façon d’appréhender le problème de la prévision est restée à une échelle locale. Pour chaque section de route, un modèle de prévision est adapté et optimisé. Notre travail de thèse présente un cadre de prévision du trafic routier qui aborde la question à l’échelle du réseau. L’étude menée au sein de ces travaux de thèse vise à exposer et évaluer cette nouvelle approche, dite globale, au regard d’approches usuelles, puis à analyser sa sensibilité vis-à-vis de divers facteurs. Après un positionnement par rapport à l’état de l’art en théorie du trafic, le cadre prédictif fondé sur des méthodes de prévision multi-variées par apprentissage est détaillé. Une version multidimensionnelle des k plus proches voisins, modèle parcimonieux et simple, est évaluée sur divers cas d’études. L’originalité réside dans l’exploitation de données issues de méthodes innovantes de collecte (e.g. Bluetooth, véhicules traceurs, véhicules connectés). Par la suite, les performances de l’approche initiale sont comparées à d’autres méthodes d’apprentissage. Un effort particulier est porté sur l’adaptation de méthodes à noyaux au cadre prédictif global. Les performances obtenues laissent entrevoir une typologie des méthodes en fonction des caractéristiques spatiotemporelles du réseau. Afin d’améliorer les performances en prévision et de réduire les temps de calcul, une méthode d’identification et de sélection des sections critiques du réseau est proposée. Les résultats prouvent qu’un sous-ensemble restreint de sections est en effet suffisant pour garantir des performances satisfaisantes en généralisation. Enfin, la résilience du cadre prédictif est évaluée au regard des événements non récurrents affectant le fonctionnement nominal du réseau, comme des incidents ou des conditions météorologiques dégradées. Les résultats soulignent l’impact de ces conditions non récurrentes sur la prévision temps-réel de la dynamique court-terme d’un réseau et permettent de dresser une feuille de route pour l’élaboration d’un cadre prédictif résilient et opérationnel. Cette nouvelle vision de la prévision s’inscrit dans les perspectives actuelles en termes d’applications sur les modules embarqués et les objectifs des gestionnaires d’infrastructures.
- 3 April 2026
- Published by Agence Bibliographique de l'Enseignement Supérieur
Contributions à la prévision court-terme, multi-échelle et multi-variée, par apprentissage statistique du trafic routier
La maturité de la télématique et des technologies de l’information et la communication (TIC), ainsi que l’avènement du big data dans le transport ont conduit à des développements foisonnants dans le domaine des systèmes de transports intelligents (ITS), aussi bien sur le plan des technologies de recueil que du traitement innovant de l’information. Il est désormais possible de connaître les conditions de circulation et les états de trafic sur la plupart des sections d’un réseau routier sans avoir recours à des infrastructures intrusives de collecte de données, de transmettre l’information résultante via des réseaux sans fil et de traiter rapidement toutes ces données multi-sources disponibles. La constitution de grandes bases de données a naturellement fait évoluer la pratique de gestion du trafic et plus particulièrement les méthodes de prévision. Ces méthodes ont connu un renouveau en s’inspirant des travaux produits en apprentissage statistique. Néanmoins, la façon d’appréhender le problème de la prévision est restée à une échelle locale. Pour chaque section de route, un modèle de prévision est adapté et optimisé. Notre travail de thèse présente un cadre de prévision du trafic routier qui aborde la question à l’échelle du réseau. L’étude menée au sein de ces travaux de thèse vise à exposer et évaluer cette nouvelle approche, dite globale, au regard d’approches usuelles, puis à analyser sa sensibilité vis-à-vis de divers facteurs. Après un positionnement par rapport à l’état de l’art en théorie du trafic, le cadre prédictif fondé sur des méthodes de prévision multi-variées par apprentissage est détaillé. Une version multidimensionnelle des k plus proches voisins, modèle parcimonieux et simple, est évaluée sur divers cas d’études. L’originalité réside dans l’exploitation de données issues de méthodes innovantes de collecte (e.g. Bluetooth, véhicules traceurs, véhicules connectés). Par la suite, les performances de l’approche initiale sont comparées à d’autres méthodes d’apprentissage. Un effort particulier est porté sur l’adaptation de méthodes à noyaux au cadre prédictif global. Les performances obtenues laissent entrevoir une typologie des méthodes en fonction des caractéristiques spatiotemporelles du réseau. Afin d’améliorer les performances en prévision et de réduire les temps de calcul, une méthode d’identification et de sélection des sections critiques du réseau est proposée. Les résultats prouvent qu’un sous-ensemble restreint de sections est en effet suffisant pour garantir des performances satisfaisantes en généralisation. Enfin, la résilience du cadre prédictif est évaluée au regard des événements non récurrents affectant le fonctionnement nominal du réseau, comme des incidents ou des conditions météorologiques dégradées. Les résultats soulignent l’impact de ces conditions non récurrentes sur la prévision temps-réel de la dynamique court-terme d’un réseau et permettent de dresser une feuille de route pour l’élaboration d’un cadre prédictif résilient et opérationnel. Cette nouvelle vision de la prévision s’inscrit dans les perspectives actuelles en termes d’applications sur les modules embarqués et les objectifs des gestionnaires d’infrastructures.
- 13 October 2025
- Published by Taylor & Francis in Journal of Intelligent Transportation Systems
- Vol. 30 (4) , 1-27
Unifying expert knowledge and field data toward an enhanced scenario description for CAV certification: a comprehensive scenario-based approach
- 13 October 2025
- Published by Taylor & Francis in Journal of Intelligent Transportation Systems
- Vol. 30 (4) , 1-27
Unifying expert knowledge and field data toward an enhanced scenario description for CAV certification: a comprehensive scenario-based approach
- 25 June 2025
- Published by Springer Nature in Data Science for Transportation
- Vol. 7 (2) , 1-23
Detection of Dangerously Slow Vehicles on Highway Corridor in a Partially Connected Environment
- 25 June 2025
- Published by Springer Nature in Data Science for Transportation
- Vol. 7 (2) , 1-23
Detection of Dangerously Slow Vehicles on Highway Corridor in a Partially Connected Environment
- 16 April 2025
- Published by Springer Nature
- p. 419-433
A Safety Kernel for Flexible Level Crossing Management
- 16 April 2025
- Published by Springer Nature
- p. 419-433
A Safety Kernel for Flexible Level Crossing Management
- 3 October 2024
- Published by Springer Nature in Data Science for Transportation
- Vol. 6 (3) , 1-22
Toward a Criticality-Guided Sampling Strategy to Reduce Tests for Automated Driving Validation
- 3 October 2024
- Published by Springer Nature in Data Science for Transportation
- Vol. 6 (3) , 1-22
Toward a Criticality-Guided Sampling Strategy to Reduce Tests for Automated Driving Validation